How generalizable is a machine-learning approach for modeling hub-height turbulence intensity? - NREL

Page created by Antonio Kelley
 
CONTINUE READING
How generalizable is a machine-learning approach for modeling hub-height turbulence intensity? - NREL
The Science of Making Torque from Wind (TORQUE 2022)                                                            IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028                            doi:10.1088/1742-6596/2265/2/022028

How generalizable is a machine-learning approach for
modeling hub-height turbulence intensity?
                      Nicola Bodini1 , Julie K. Lundquist1,2,3 , Hannah Livingston2 and
                      Patrick Moriarty1
                      1
                          National Renewable Energy Laboratory, Golden, Colorado 80401, USA
                      2
                          University of Colorado Boulder, Boulder, Colorado 80302, USA
                      3
                          Renewable and Sustainable Energy Institute, Boulder, Colorado 80303, USA
                      E-mail: nicola.bodini@nrel.gov

                      Abstract. Hub-height turbulence intensity is essential for a variety of wind energy
                      applications. However, simulating it is a challenging task. Simple analytical models have been
                      proposed in the literature, but they all come with significant limitations. Even state-of-the-art
                      numerical weather prediction models, such as the Weather Research and Forecasting model,
                      currently struggle to predict hub-height turbulence intensity. Here, we propose a machine-
                      learning-based approach to predict hub-height turbulence intensity from other hub-height and
                      ground-level atmospheric measurements, using observations from the Perdigão field campaign
                      and the Southern Great Plains atmospheric observatory. We consider a random forest regression
                      model, which we validate first at the site used for training and then under a more robust round-
                      robin approach, and compare its performance to a multivariate linear regression. The random
                      forest successfully outperforms the linear regression in modeling hub-height turbulence intensity,
                      with a normalized root-mean-square error as low as 0.014 when using 30-minute average data. In
                      order to achieve such low root-mean-square error values, the knowledge of hub-height turbulence
                      kinetic energy (which can instead be modeled in the Weather Research and Forecasting model) is
                      needed. Interestingly, we find that the performance of the random forest generalizes well when
                      considering a round-robin validation (i.e., when the algorithm is trained at one site such as
                      Perdigão or Southern Great Plains) and then applied to model hub-height turbulence intensity
                      at the other location.

1. Introduction
As wind energy expands and becomes an increasingly large portion of energy portfolios, collecting
data on atmospheric turbulence at turbine hub height is essential for optimizing the performance
of wind power plants and their successful integration into the electric system. Hub-height
turbulence intensity (TI) impacts turbine power generation [1, 2, 3, 4, 5, 6], and is instrumental
in the control strategy for wake modeling and wake-steering strategies [7, 8]. Turbine siting and
power forecasting also rely on hub-height turbulence and are determining factors of whether a
wind power plant will be financially advantageous [9, 10].
   Various formulations have been proposed for estimating hub-height TI, or TI profiles, as a
function of surface quantities. The Mann model [11, 12] provided an analytic characterization
of turbulence profiles in the presence of uniform linear shear. Cheung et al. [13] proposed
and tested a formulation based on assumptions for turbulent boundary layers in a range of
stability conditions. Building from the power-law approach for extrapolating wind speeds from

              Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
              of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd                          1
How generalizable is a machine-learning approach for modeling hub-height turbulence intensity? - NREL
The Science of Making Torque from Wind (TORQUE 2022)                                    IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028    doi:10.1088/1742-6596/2265/2/022028

surface-level measurements to hub height, Gualtieri [14] proposed and tested an empirically
based approach for extrapolating 30-m measurements to 100-m altitudes and found that it is
critical to consider atmospheric stability. Türk and Emeis [15] showed how offshore turbulence
intensity has a strong dependence on wind speed and wave height, and proved that specifications
given by the International Electrotechnical Commission (IEC) wind turbine standards 61400-1
[16] and 61400-3 [17] are inaccurate for their test location. All these efforts come with specific
limitations in their range of applicability, so a more universal model for hub-height TI is currently
not available.
   Further, numerical weather prediction models do not routinely predict TI. For example, the
state-of-the-art numerical weather prediction model, the Weather Research and Forecasting
(WRF) model, [18, 19]), does not provide TI as output. Only some choices of planetary
boundary layer schemes (i.e., the MYNN scheme [20, 21]) even provide a prognostic calculation
of turbulence kinetic energy (TKE), while others widely used for wind energy applications (i.e.,
the YSU scheme [22]) do not even estimate TKE. And in any case, the spatial scales at which
numerical weather prediction models are run are too coarse for some applications, such as turbine
control, where the knowledge of hub-height TI is essential.
   Given the complexity and inherent nonlinearity of atmospheric turbulence, machine-learning
techniques could be leveraged for the derivation of a model for hub-height TI. Machine-learning
approaches have been successfully applied to several wind energy applications, ranging from
power forecasting [23, 9] to issues at the wind turbine scale, and for turbine power curve modeling
[24], turbine faults and controls [25], and turbine blade management [26]. Recently, machine
learning has also been successfully applied to vertically extrapolate wind speed [27, 28, 29],
both on land and offshore, to better represent the turbulence dissipation rate [30] and to better
parameterize surface fluxes [31, 32].
   In this analysis, we develop and perform a thorough validation of a machine-learning approach
to model hub-height TI, calculated as a 30-minute average as:
                                                      σU
                                             TI =        ,                                        (1)
                                                      U

where σU is the standard deviation of the horizontal wind speed, and U is the 30-minute average
horizontal wind speed. We consider observations from the Perdigão field campaign [33] and
the U.S. Department of Energy Atmospheric Radiation Measurement Southern Great Plains
(SGP) observatory [34], which are both presented in Section 2, together with a description of
the proposed machine-learning method. In Section 3, we first derive physical insights on the
variability of hub-height TI from the application of the proposed approach at a single site. Next,
we test the generalization accuracy of the model when adopting a more practical validation setup,
which resembles the use of the WRF model, where the algorithm is trained at one location and
then applied to model hub-height TI at a different site. Finally, we summarize our results and
suggest future work in Section 4.

2. Data and methods
2.1. Observational data sets
The first set of observations we consider is taken from the Perdigão field campaign [33], an
international effort that brought scientists from multiple institutions to a valley in central
Portugal (see map in figure 1a), where an intensive observational period was completed from 1
May 2017 to 15 June 2017. The site is characterized by a complex topography, with a difference
in altitude of about 200 m from the bottom of the valley to the top of the two almost-parallel
ridges, which are separated by 1.5 km. The site is largely covered in trees, which increase the
surface roughness. The areas outside the ridges and valley are largely farmland and eucalyptus
groves. For this analysis, we use data from three 100-m meteorological towers (tse04, tse09,

                                                  2
The Science of Making Torque from Wind (TORQUE 2022)                                   IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028   doi:10.1088/1742-6596/2265/2/022028

(a)                                                  (b)

Figure 1. Map of (a) the Perdigão field campaign and (b) SGP locations, with the locations
of the meteorological towers considered in this study.

and tse13—see map in figure 1a). At each tower, we consider observations from the 20-Hz sonic
anemometers at 10 m and 60 m above ground level (AGL). To exclude tower wake effects, we
discard from analysis those time stamps when the wind direction was ±30◦ from the direction
of the tower boom. We also discard periods of precipitation to avoid data contamination [35],
and periods when the 60-m wind speed was lower than 3 m s−1 , which are not relevant for wind
energy purposes.
   Next, we use observations from the main meteorological tower at the SGP observatory [34] in
northern Oklahoma in the United States. The rural region is characterized by relatively simple
topography, and its land use is primarily cattle pasture and wheat fields (figure 1b). Observations
are from sonic anemometers at 10 m and 60 m AGL, publicly available as 30-minute averages,
and span from 22 July 2015 to 15 March 2022.

2.2. Regression algorithms
In our analysis, we consider two regression algorithms, and we will compare their performance in
predicting 30-minute average TI at 60 m AGL. For both regression models, we use as inputs the
following variables: wind speed at 10 m AGL, TKE at 10 m AGL, TKE normalized by wind speed
(TKE/WS) at 10 m AGL, Obukhov length (L) at 10 m AGL, proxy for atmospheric stability,
wind speed at 60 m AGL, TKE at 60 m AGL, TKE normalized by wind speed (TKE/WS) at
60 m AGL. Although obviously correlated with TKE and wind speed, we decide to include a
normalized version of TKE as an additional input because it mimics the functional structure
of TI, where horizontal wind speed is at the denominator. As already mentioned, we discard
from the analysis all time stamps where 60-m wind speed is lower than a typical cut-in wind
speed for modern wind turbines, here set to 3 m s−1 , a wind regime not relevant for wind energy
purposes.
    As a baseline, we consider a multivariate linear regression. To avoid training a model that
overfits the data, regularization techniques need to be implemented so that the learning model
is constrained: the fewer degrees of freedom the model has, the harder it will be for it to overfit
the data. We use Ridge regression [36] (Ridge in Python’s library scikit-learn) to constrain the
multivariate regression. The Ridge regression is achieved by adding a regularization term to
the cost function (mean square error), with a hyperparameter α that controls how much the
model will be regularized. The optimal value of α is determined by cross validation, with values
sampled in the range from 0.1 to 10.

                                                 3
The Science of Making Torque from Wind (TORQUE 2022)                                     IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028     doi:10.1088/1742-6596/2265/2/022028

    Next, as a proof of concept of the skills of more sophisticated machine-learning algorithms, we
use a random forest regression model (RandomForestRegressor module in Python’s scikit-learn
[37]). We note here that ensemble models such as the random forest have shown strong predictive
power in previous work [9, 27], and we defer a more exhaustive analysis of the predictive skills
of other learning algorithms to a later study. The set of algorithm hyperparameters considered
in the training and validation process and their sampled ranges used in the cross validation are
listed in table 1.

Table 1. Algorithm hyperparameters considered for the random forest and their considered
values in the cross validation.
                               Hyperparameter                    Sampled Values
                          Number of estimators                       10–800
                            Maximum depth                             4–40
                       Maximum number of features                      1–7
                   Minimum number of samples to split                 2–11
                   Minimum number of samples for a leaf               1–15

   For each regression algorithm, we adopt a nested cross-validation approach. We divide the
data set into three subsets: the training, validation, and testing sets [38]. First, the learning
algorithm is trained multiple times with different hyperparameters on the training set. Second,
the best set of hyperparameters is selected as the one that gives the best results when applying
the model on the validation set. Finally, the generalization skills of the learning model are
evaluated on the testing set, which was not seen by the model during the optimization of the
regression algorithm. Given the time-series nature of our problem, the construction of the
training, validation, and testing sets needs to be carefully considered. If the data in each of the
three sets were picked randomly, the performance of the machine-learning model would likely
be artificially improved. In fact, if created randomly, we could end up in a situation where the
observations at a given time stamp would be in the training set, and the observations at the
subsequent time stamp, which are likely highly correlated with those from the previous one,
could instead be part of the validation or testing set. Therefore, we use contiguous data to build
the three sets: we keep a contiguous 20% of the data for testing, whereas the remaining 80%
are used in a five-fold cross validation to optimize the model hyperparameters. To obtain the
best possible estimate of the generalization skill of the model, in this nested approach we repeat
the training process by shifting the testing set five times, so that it can cover the full period
of record. We therefore evaluate the overall performance based on the root-mean-square error
(RMSE) between the actual and predicted hub-height TI, averaged over the five different testing
periods.

3. Results
We first consider the predictive skills of the multivariate linear regression approach when it
is trained and tested at the same site. Figure 2 shows scatterplots of observed and linear-
regression-predicted (over the testing set) TI at 60 m AGL at the three Perdigão towers and
the SGP main tower. We quantify the skills of the considered regression algorithm in terms of
the RMSE normalized by the range of observed TI values, to be able to make a fair comparison
among different sites. The results show NRMSE between 0.067 and 0.086 at the Perdigão towers
and 0.020 at SGP.

                                                 4
The Science of Making Torque from Wind (TORQUE 2022)                                   IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028   doi:10.1088/1742-6596/2265/2/022028

    (a)          Perdigão - tse04             (b)           Perdigão - tse09

    (c)                                       (d)
                 Perdigão – tse13                                 SGP

Figure 2. Scatterplots of observed and linear-regression-predicted 30-minute average TI at 60
m AGL at (a–c) the three Perdigão towers and (d) the SGP tower, when the regression model
is trained and tested at the same location.

    We then compare these baseline results to what can be achieved when adopting the more
sophisticated random forest approach. We find good performance at all sites, with NRMSE
ranging from 0.012 (in simple terrain and with a much bigger training set) to 0.071 (in the very
complex terrain of the bottom of the Perdigão valley, Tower tse09). At Perdigão, these results
represent a reduction of 18-25% compared to the multivariate linear regression results for the
two towers on the ridges, and a reduction of just 6% for tse09, where the reduced number of
available data and the impact of the ultra-complex terrain and canopy make it a challenging site
to model, even for a more sophisticated random forest algorithm. At SGP, where the terrain
is simpler and the training period includes multiple years of observations, we observe a 30%
reduction in NRMSE when moving from a linear regression to the random forest approach.
    Next, we assess the importance of the various input variables used to feed the random forest.
First, we determine how the testing NRMSE varies as the random forest is fed with more and
more input variables (figure 4). We start by training a random forest with 10-m wind speed
as the only input, and then we progressively add more near-surface input variables before also
adding hub-height quantities. In general, using near-surface wind speed as the only input does
not allow for an accurate modeling of hub-height TI. When including near-surface turbulence
and stability metrics as inputs, the NRMSE drops, but it still remains significantly larger than
the values seen when including variables observed at hub-height as additional inputs. Finally, if

                                                 5
The Science of Making Torque from Wind (TORQUE 2022)                                   IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028   doi:10.1088/1742-6596/2265/2/022028

    (a)          Perdigão - tse04             (b)           Perdigão - tse09

    (c)                                       (d)
                 Perdigão – tse13                                 SGP

Figure 3. Scatterplots of observed and random-forest-predicted 30-minute average TI at 60 m
AGL at the three Perdigão towers and the SGP one, when the regression model is trained and
tested at the same location, for the testing set.

no data on hub-height turbulence are given to the model, and wind speed is the only hub-height
input feature, we still find good performance, with testing NRMSE less than 0.1, but still higher
than the lowest values found with the full set of inputs, especially in simple terrain at SGP.
   As a way to summarize this analysis, we report in figure 5 the relative feature importance of
each input variable at the Perdigão tse09 and SGP towers (results from the two other towers at
Perdigão were similar). This metric is calculated by assessing to what degree the random forest
tree nodes that use each feature reduce the mean square error on average (across all trees in the
forest), weighted by the number of times each feature is selected. At both locations, we see how
hub-height variables are the most influential in predicting hub-height TI, which is consistent
with the results of the analysis in figure 4. Still, near-surface variables help further improve
the accuracy of the random forest, especially at the Perdigão site. Although useful, we note
how this metric is affected by the existing correlation between different inputs: for example, if
the 60-m normalized TKE was not included as input, we would likely see a significantly higher
importance for the other two hub-height variables (wind speed and TKE).
   Finally, we leverage the data sets from both locations to understand the generalization skills
of the proposed random forest approach. Specifically, we apply a round-robin validation, where
we train the learning model at one site (either SGP or Perdigão), and then apply it to model
60-m TI at the other location, and vice versa. Figure 6 shows the results of this validation
exercise in terms of scatterplots of observed and machine-learning-predicted 60-m TI. While the

                                                    6
The Science of Making Torque from Wind (TORQUE 2022)                                       IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028       doi:10.1088/1742-6596/2265/2/022028

                                           Testing NRMSE

 tse04     0.14        0.11        0.12         0.1          0.077          0.056       0.055

 tse09     0.17        0.12        0.11        0.11          0.095          0.073       0.071

 tse13     0.17        0.13        0.13        0.12          0.091          0.067       0.065

  SGP      0.044       0.029       0.029       0.028         0.024          0.014       0.012

         WS 10m     +TKE 10m +TKE/WS 10m +L 10m            +WS 60m       +TKE 60m +TKE/WS 60m

Figure 4. Testing NRMSE when a random forest is used to model 30-minute average 60-m TI
at the Perdigão and SGP towers, when multiple variables are progressively added as inputs to
the learning algorithm.

NRMSE values are larger than the optimal ones found for a same-site exercise (see figure 3),
we still see a remarkable accuracy for both cases, with NRMSE values of 0.03 and 0.06 when
the model is applied at SGP and Perdigão, respectively. We find a larger increase, compared to
the same-site scenario, when training the model at Perdigão and applying it at SGP, possibly
because of the very limited length of the training data at Perdigão. On the other hand, when
training the model at SGP and applying it at Perdigão, the NRMSE is not significantly different
than what observed in figure 3, under the same-site scenario.

4. Conclusions
The ability to model hub-height turbulence intensity would be extremely useful in all applications
where hub-height turbulence is crucial, ranging from wind turbine control to power forecasting
and—in a broader perspective—pollutant dispersion and drone flight forecasting. Such a model,
if incorporated in the state-of-the-art numerical weather prediction model at the mesoscale
(WRF), would represent a huge advancement for the wind energy modeling community. Despite
this intense motivation to find accurate approaches for estimating turbulence conditions at hub
height, few models have been proposed to date for such application, each with their inherent
uncertainty and limitations. The challenge is even larger for sites in complex terrain.
    We analyzed atmospheric data at 10 m and 60 m above ground level from four meteorological
towers at two sites: three towers from the Perdigão field campaign, and a fourth tower at the
SGP site. We considered how a machine-learning-based algorithm could be derived and applied
to successfully model hub-height TI. We focused our validation efforts first using a same-site
approach, where the random forest is trained and then tested at the same site, which is useful to
derive several physical insights into the proposed model. We find that the proposed random
forest is successful in modeling hub-height TI, with testing NRMSE as low as 0.014 when
applied on 30-minute average data. In order to obtain such low values of testing NRMSE,
the knowledge of hub-height TKE (which can instead be modeled in WRF) is needed. Next,
we considered a more practically useful (and scientifically fair) round-robin validation, where

                                                  7
The Science of Making Torque from Wind (TORQUE 2022)                                   IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028   doi:10.1088/1742-6596/2265/2/022028

               (a) Perdigão – tse09

                (b) SGP

Figure 5. Relative importance of the various input variables used to feed the random forest in
predicting 10-minute average 60-m TI at (a) the Perdigão tse09 tower and (b) the SGP tower.

the random forest is trained at one site and then applied to predict hub-height TI at a different
site. Notably, we find a good generalization performance of our proposed model even under this
scenario, especially when the training set includes many years of data (in our case, at SGP).
Our validation results confirm the good generalization skills of the proposed algorithm, despite
the significant differences (in terrain, canopy, and duration of the period of record) between the
two considered field campaigns.
    While promising, several opportunities exist to further expand our work in the future. A more
complete survey of additional learning algorithms and/or input variables could be completed to
optimize the performance of the machine-learning approach. Also, one could expand the training
and testing validation by considering different sites, such as those offshore. Finally, this more
complete and broadly validated model could be incorporated in WRF to add to it a capability
to output TI.

Acknowledgments
This work was authored in part by the National Renewable Energy Laboratory, operated by
Alliance for Sustainable Energy, LLC, for the U.S. Department of Energy (DOE) under Contract
No. DE-AC36-08GO28308. Funding provided by the U.S. Department of Energy Office of
Energy Efficiency and Renewable Energy Wind Energy Technologies Office. The views expressed
in the article do not necessarily represent the views of the DOE or the U.S. Government. The U.S.

                                                 8
The Science of Making Torque from Wind (TORQUE 2022)                                         IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028         doi:10.1088/1742-6596/2265/2/022028

                            (a)

 (b)            Train at Perdigão, test                (c)                  Train at SGP,
                        at SGP                                            test at Perdigão

Figure 6. (a) Histograms of 30-minute average TI at 60 m AGL for the three Perdigão towers
combined and the SGP one. (b) Scatterplot of observed and random-forest-predicted 30-minute
average 60-m TI when the random forest is trained at the three Perdigão towers combined and
tested at SGP. (c) Same as (b) but with training and tested sets switched.

Government retains and the publisher, by accepting the article for publication, acknowledges
that the U.S. Government retains a nonexclusive, paid-up, irrevocable, worldwide license to
publish or reproduce the published form of this work, or allow others to do so, for U.S.
Government purposes.
We thank the residents of Alvaiade and Vale do Cobrão for their essential hospitality and support
throughout the Perdigão field campaign. In particular, we are grateful for the human and logistic
support Felicity Townsend provided to our research group in the field. We express appreciation
to the Earth Observing Laboratory at the National Center for Atmospheric Research, the Danish
Technical University, and INEGI for their herculean efforts to construct and maintain the data
streams from the networks of towers at Perdigão.

References
 [1] Elliott D L and Cadogan J B 1990 Effects of wind shear and turbulence on wind turbine power curves Tech.
       rep. Pacific Northwest Lab., Richland, WA (USA)

                                                      9
The Science of Making Torque from Wind (TORQUE 2022)                                          IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028          doi:10.1088/1742-6596/2265/2/022028

 [2] Wharton S and Lundquist J K 2012 Wind Energy 15 525–546 ISSN 1095-4244 URL
        https://onlinelibrary.wiley.com/doi/abs/10.1002/we.483
 [3] Wharton S and Lundquist J K 2012 Environmental Research Letters 7 014005 ISSN 1748-9326
 [4] Vanderwende B J and Lundquist J K 2012 Environmental Research Letters 7 034035 ISSN 1748-9326
 [5] Murphy P, Lundquist J K and Fleming P 2020 Wind Energy Science 5 1169–1190 ISSN 2366-7443 publisher:
        Copernicus GmbH URL https://wes.copernicus.org/articles/5/1169/2020/
 [6] Vahidzadeh      M     and    Markfort    C     D     2019    Energies     12    ISSN    1996-1073      URL
        https://www.mdpi.com/1996-1073/12/9/1805
 [7] Fleming P, King J, Dykes K, Simley E, Roadman J, Scholbrock A, Murphy P, Lundquist J K, Moriarty
        P, Fleming K, van Dam J, Bay C, Mudafort R, Lopez H, Skopek J, Scott M, Ryan B, Guernsey C
        and Brake D 2019 Wind Energy Science 4 273–285 ISSN 2366-7443 publisher: Copernicus GmbH URL
        https://wes.copernicus.org/articles/4/273/2019/
 [8] Fleming P, King J, Simley E, Roadman J, Scholbrock A, Murphy P, Lundquist J K, Moriarty P,
        Fleming K, van Dam J, Bay C, Mudafort R, Jager D, Skopek J, Scott M, Ryan B, Guernsey C and
        Brake D 2020 Wind Energy Science 5 945–958 ISSN 2366-7443 publisher: Copernicus GmbH URL
        https://wes.copernicus.org/articles/5/945/2020/
 [9] Optis M and Perr-Sauer J 2019 Renewable and Sustainable Energy Reviews 112 27–41
[10] Bodini N, Optis M, Perr-Sauer J, Simley E and Fields M J 2021 Wind Energy
[11] Mann J 1994 Journal of Fluid Mechanics 273 141–168 ISSN 1469-7645, 0022-1120 publisher: Cambridge
        University Press
[12] Mann J 1998 Probabilistic Engineering Mechanics 13 269–282 ISSN 0266-8920 URL
        https://www.sciencedirect.com/science/article/pii/S0266892097000362
[13] Cheung      L   C,    Premasuthan     S,   Davoust     S   and      von   Terzi   D   2016     Journal of
        Physics:        Conference    Series     753     032008      ISSN     1742-6588,     1742-6596      URL
        https://iopscience.iop.org/article/10.1088/1742-6596/753/3/032008
[14] Gualtieri     G     2017      Renewable      Energy      111      175–186     ISSN     0960-1481       URL
        https://www.sciencedirect.com/science/article/pii/S0960148117302872
[15] Türk M and Emeis S 2010 Journal of Wind Engineering and Industrial Aerodynamics 98 466–471
[16] 61400-1 I 2019 Wind turbines—part 3: design requirements
[17] 61400-3 I 2019 Wind turbines—part 1: design requirements for offshore wind turbines
[18] Powers J G, Klemp J B, Skamarock W C, Davis C A, Dudhia J, Gill D O, Coen J L, Gochis D J, Ahmadov
        R, Peckham S E et al. 2017 Bulletin of the American Meteorological Society 98 1717–1737
[19] Skamarock W C, Klemp J B, Dudhia J, Gill D O, Liu Z, Berner J, Wang W, Powers J G, Duda M G, Barker
        D M et al. 2019 National Center for Atmospheric Research: Boulder, CO, USA 145
[20] Nakanishi M and Niino H 2009 Journal of the Meteorological Society of Japan. Ser. II 87 895–912
[21] Olson J B, Kenyon J S, Angevine W, Brown J M, Pagowski M, Sušelj K et al. 2019
[22] Hu X M, Klein P M and Xue M 2013 Journal of Geophysical Research: Atmospheres 118 10–490
[23] Giebel G, Brownsword R, Kariniotakis G, Denhard M and Draxl C 2011 ANEMOS. plus
[24] Clifton A, Kilcher L, Lundquist J and Fleming P 2013 Environmental research letters 8 024009
[25] Leahy K, Hu R L, Konstantakopoulos I C, Spanos C J and Agogino A M 2016 2016 IEEE International
        Conference on Prognostics and Health Management (ICPHM) (IEEE) pp 1–8
[26] Arcos Jiménez A, Gómez Muñoz C and Garcı́a Márquez F 2018 Energies 11 13
[27] Bodini N and Optis M 2020 Wind Energy Science 5 489–501
[28] Bodini N and Optis M 2020 Journal of Physics: Conference Series vol 1618 (IOP Publishing) p 062037
[29] Optis M, Bodini N, Debnath M and Doubrawa P 2021 Wind Energy Science Discussions 1–26
[30] Bodini N, Lundquist J K and Optis M 2020 Geoscientific Model Development 13 4271–4285
[31] Leufen L H and Schädler G 2019 Geoscientific model development 12 2033–2047
[32] McCandless T, Gagne D, Kosovic B, Haupt S E, Yang B, Becker C and Schreck J in review Boundary Layer
        Meteorology
[33] Fernando H J S, Mann J, Palma J M L M, Lundquist J K, Barthelmie R J, Belo-Pereira M, Brown
        W O J, Chow F K, Gerz T, Hocut C M, Klein P M, Leo L S, Matos J C, Oncley S P, Pryor S C,
        Bariteau L, Bell T M, Bodini N, Carney M B, Courtney M S, Creegan E D, Dimitrova R, Gomes
        S, Hagen M, Hyde J O, Kigle S, Krishnamurthy R, Lopes J C, Mazzaro L, Neher J M T, Menke R,
        Murphy P, Oswald L, Otarola-Bustos S, Pattantyus A K, Rodrigues C V, Schady A, Sirin N, Spuler S,
        Svensson E, Tomaszewski J, Turner D D, Veen L v, Vasiljević N, Vassallo D, Voss S, Wildmann N and
        Wang Y 2019 Bulletin of the American Meteorological Society 100 799–819 ISSN 0003-0007, 1520-0477
        publisher: American Meteorological Society Section: Bulletin of the American Meteorological Society URL
        https://journals.ametsoc.org/view/journals/bams/100/5/bams-d-17-0227.1.xml
[34] Sisterson D, Peppler R, Cress T, Lamb P and Turner D 2016 Meteorological Monographs 57 6–1

                                                       10
The Science of Making Torque from Wind (TORQUE 2022)                                         IOP Publishing
Journal of Physics: Conference Series       2265 (2022) 022028         doi:10.1088/1742-6596/2265/2/022028

[35] Zhang R, Huang J, Wang X, Zhang J A and Huang F 2016 Journal of Ocean University of China 15 389–398
[36] Hoerl A E and Kennard R W 1970 Technometrics 12 55–67
[37] Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss
       R, Dubourg V et al. 2011
[38] Friedman J, Hastie T and Tibshirani R 2001 The elements of statistical learning vol 1 (Springer series in
       statistics New York)

                                                      11
You can also read